Christian Weber

Abstract

In manufacturing environments, machine learning models are being built for several use cases, such as predictive maintenance and product quality control. In this context, the various manufacturing processes, machines, and product variants make it necessary to create and use lots of different machine learning models. This calls for a software system that is able to manage all these diverse machine learning models and associated metadata. However, current model management systems do not associate models with business and domain context to provide non-expert users with tailored functions for model search and discovery. Moreover, none of the existing systems provides a comprehensive overview of all models within an organization. In our demonstration, we present the MMP, our model management platform that addresses these issues. The MMP provides a model metadata extractor, a model registry, and a context manager to store model metadata in a central metadata store. On top of this, the MMP provides frontend components that offer the above-mentioned functionalities that are missing in existing systems and approaches. In our demonstration, we show two scenarios for model management in Industry 4.0 environments that illustrate the novel functionalities of the MMP. We demonstrate to the audience how the platform and its metadata, linking models to their business and domain context, help non-expert users to search and discover models. Furthermore, we show how to use MMP's powerful visualizations for model reporting, such as a dashboard and a model landscape view.

Abstract

Industry 4.0 use cases such as predictive maintenance and product quality control make it necessary to create, use and maintain a multitude of different machine learning models. In this setting, model management systems help to organize models. However, concepts for model management systems currently focus on data scientists, but do not support non-expert users such as domain experts and business analysts. Thus, it is difficult for them to reuse existing models for their use cases. In this paper, we address these challenges and present an architecture, a metadata schema and a corresponding model management platform.

Abstract

Machine Learning (ML) applications are becoming increasingly valuable due to the rise of IoT technologies. That is, sensors continuously gather data from different domains and make them available to ML for learning its models. This provides profound insights into the data and enables predictions about future trends. While ML has many advantages, it also represents an immense privacy risk. Data protection regulations such as the GDPR address such privacy concerns, but practical solutions for the technical enforcement of these laws are also required. Therefore, we introduce AMNESIA, a privacy-aware machine learning model provisioning platform. AMNESIA is a holistic approach covering all stages from data acquisition to model provisioning. This enables to control which application may use which data for ML as well as to make models “forget” certain knowledge.

Abstract

The management of machine learning models is an extremely challenging task. Hundreds of prototypical models are being built and just a few are mature enough to be deployed into operational enterprise information systems. The lifecycle of a model includes an experimental phase in which a model is planned, built and tested. After that, the model enters the operational phase that includes deploying, using, and retiring it. The experimental phase is well known through established process models like CRISP-DM or KDD. However, these models do not detail on the interaction between the experimental and the operational phase of machine learning models. In this paper, we provide a new process model to show the interaction points of the experimental and operational phase of a machine learning model. For each step of our process, we discuss according functions which are relevant to managing machine learning models.

Abstract

Historically evolved decentral organisation structures led to suboptimal implemented improvement measures by neglecting comprehensive dependencies within the tool life cycle. Cost saving measures in one life cycle process can manifest as performance deficits in others, for instance the interdependencies of low priced cutting tools and the resulting quality loss within the cutting tool operation. Therefore, a life cycle-orientated indicator is necessary to support the holistic evaluation of planned or implemented measures and their impact towards the entire tool life cycle system. This paper presents the hierarchical tool total costs of ownership (TTCO) indicator which considers tool-related costs and tool unit performance parameters. Furthermore, the indicator supports the assessment of long-term measure impacts by calculating the TTCO-based experience curve. An IT architecture concept was developed to support the time-intense (TTCO) data acquisition, calculation and its analysis towards implemented organisational and technical improvement measures by providing assisting applications.

Abstract

Für produzierende Unternehmen stellt die effiziente Verarbeitung großer Datenmengen eine Herausforderung dar. Die Auswahl der richtigen Architekturkonzepte für IT-Lösungen zur Datenverarbeitung spielt dabei eine wichtige Rolle. Um die IT an den Herausforderungen von Industrie 4.0 auszurichten, stehen Unternehmen verschiedene Referenzarchitekturen internationaler Gremien zur Verfügung. Die Hauptbeiträge dieses Artikels haben das Ziel, (i) einen Überblick über die wichtigsten Referenzarchitekturen für Industrie 4.0 (I4.0) zu geben und (ii) diese unter dem Aspekt der Datenverarbeitung zu untersuchen. Dazu werden die Referenzarchitekturen anhand von Datenverarbeitungsanforderungen für I4.0 betrachtet. Die Untersuchung zeigt, dass die I4.0-Referenzarchitekturen jeweils einen Teilbereich der Anforderungen abdecken und sich die Konzepte gegenseitig ergänzen. (iii) Darauf aufbauend werden aus den Datenverarbeitungsanforderungen technische Konsequenzen abgeleitet und Architekturkonzepte für die Realisierung einer IT-Architektur für die Datenverarbeitung vorgestellt. Dadurch wird es IT-Architekten ermöglicht, einen Vergleich der Referenzarchitekturen hinsichtlich projektbezogener Anforderungen an die Datenverarbeitung vorzunehmen sowie geeignete Architekturentscheidungen zu treffen.

Abstract

Die Verarbeitung großer Datenmengen sowie die Wichtigkeit von Datenanalysen sind in den produzierenden Unternehmen mittlerweile angekommen. Bekannte Anwendungsbeispiele sind Digital Mock-Up in der Produktentwicklung oder Prozessoptimierung durch Predictive Maintenance. Die in letzter Zeit entwickelten Referenzarchitekturen in diesen breitgefächerten Themenfeldern betrachten dementsprechend verschiedene Aspekte in unterschiedlichen Ausprägungen. Dieser aus zwei Teilen bestehende Beitrag rekapituliert und bewertet diese Entwicklungen, um Unternehmen bei der Umsetzung ihrer eigenen individuellen Architektur Hilfestellung zu geben. Im ersten Teil wurden aktuelle Referenzarchitekturen mit ihren Architekturbausteinen im Bereich Industrie 4.0 vorgestellt. In diesem Teil werden die Referenzarchitekturen unter dem Gesichtspunkt der Themenfelder Analytics sowie Datenmanagement untersucht und bewertet.

Abstract

Die Verarbeitung großer Datenmengen sowie die Wichtigkeit von Datenanalysen sind in den produzierenden Unternehmen mittlerweile angekommen. Bekannte Anwendungsbeispiele sind Digital Mock-Up in der Produktentwicklung oder Prozessoptimierung durch Predictive Maintenance. Die in letzter Zeit entwickelten Referenzarchitekturen in diesen breitgefächerten Themenfeldern betrachten dementsprechend verschiedene Aspekte in unterschiedlichen Ausprägungen. Dieser aus zwei Teilen bestehende Beitrag rekapituliert und bewertet diese Entwicklungen, um Unternehmen bei der Umsetzung ihrer eigenen individuellen Architektur Hilfestellung zu geben. In diesem Teil werden aktuelle Referenzarchitekturen mit ihren Architekturbausteinen im Bereich Industrie 4.0 vorgestellt. Im zweiten Teil werden die Referenzarchitekturen unter dem Gesichtspunkt der Themenfelder Analytics sowie Datenmanagement untersucht und bewertet.

Abstract

To stay competitive in a global market, companies strive to adjust their IT architectures to the characteristics of Industrie 4.0. This means the adaptation of a data-driven manufacturing model, which includes the horizontal and vertical integration of the value chain and the management of data across the whole product life cycle. Currently companies face two major problems in this transformation of their IT architectures: 1) There are multiple reference architectures for Industrie 4.0, whose suitability has not been evaluated in a straightforward way. 2) There are no maturity models to objectively assess data-driven IT architectures. In this work we address these problems by providing an overview of Industrie 4.0 features in multiple reference architectures and developing a maturity model for IT architectures for data-driven manufacturing. We first give an overview of the current maturity models with aspects of data-driven manufacturing. Based on this overview we survey three reference architectures for Industrie 4.0: the Industrial Internet Reference Architecture, the Reference Architecture Model Industrie 4.0, and the Stuttgart IT Architecture for Manufacturing. We analyze and assess each approach and define maturity levels for data-driven manufacturing to guide companies in the adaption process of Industrie 4.0, as well as criteria for meeting these levels.

Abstract

The global conditions for manufacturing are rapidly changing towards shorter product life cycles, more complexity and more turbulence. The manufacturing industry must meet the demands of this shifting environment and the increased global competition by ensuring high product quality, continuous improvement of processes and increasingly flexible organization. Technological developments towards smart manufacturing create big industrial data which needs to be leveraged for competitive advantages. We present a novel IT architecture for datadriven manufacturing, the Stuttgart IT Architecture for Manufacturing (SITAM). It addresses the weaknesses of traditional manufacturing IT by providing IT systems integration, holistic data analytics and mobile information provisioning. The SITAM surpasses competing reference architectures for smart manufacturing because it has a strong focus on analytics and mobile integration of human workers into the smart production environment and because it includes concrete recommendations for technologies to implement it, thus filling a granularity gap between conceptual and case-based architectures. To illustrate the benefits of the SITAM’s prototypical implementation, we present an application scenario for value-added services in the automotive industry.